Companies rein in AI usage as costs strain budgets

People with direct experience said the cost shock is real. Some reported budgets being cut multiple times in a few months, usage caps dropping to around $500 per person, and audits for heavy users. The strongest practical point was that the spend is easy to justify in demos and hard to justify in shipped product. Teams can clear more tickets, generate more docs, and do more visible activity without moving the critical path or the bottom line. In the worst case, they create a review bottleneck, push features faster than the team can validate them, and leave behind code no one fully understands. The mood was mostly anti-hype, but not anti-tool. Several engineers said LLMs are plainly useful for narrow tasks like navigating unfamiliar APIs or accelerating coding work when the user already knows what good looks like. The consensus landing point was much narrower than the marketing pitch. AI is useful enough that large companies will keep it, but expensive enough that the free-spending phase is over. What survives is likely tighter budgeting, more scrutiny on where it actually helps, and a split between big firms that can justify dedicated infrastructure and everyone else who has to live within token caps.